AI Snowflake: Stop Asking the Wrong AI Question

One of the most common ways projects go sideways is by asking a question that AI cannot answer—or worse, one that doesn’t actually deliver ROI. (See Chapter 1 of my UX for AI Book How to Screw Up Your Project for a whole catalog of “badly formed questions” 😉)

Take a simple scheduling problem in a business context. It’s tempting to think AI could figure out something like:

• “When is the optimal time to run this meeting?”

• “What’s the best way to staff this shift?”

Those sound clever, but unless you already have complete calendar and availability data, they’re nearly impossible to answer in a useful way. Even if you did, you might not need AI at all—some of these can be solved with straightforward rules or linear regression (see my post on forecasting with line graphs).

The real ROI comes from reframing the question.

Instead of trying to predict a “nice-to-have,” go after the problem that hits revenue, efficiency, or retention directly.

For example, in a workplace context, the killer AI question might be:

“How likely is this employee to quit?”

That’s an excellent AI target because:

• You already have the data. Performance reviews, engagement scores, overtime hours, tenure, and absenteeism—all can feed the model.

• It doesn’t require extra UI plumbing. The analysis runs behind the scenes; no new scheduling widgets or tricky integrations needed.

• Your data is your moat. Nobody else has the same combination of organizational history and workforce behavior.

• It goes to the heart of the problem. Employee churn costs companies massive amounts of money.

• It’s hard for humans to see at scale. Managers may know who’s struggling, but they can’t track patterns across thousands of employees.

• It’s a proven use case. Predicting attrition is already widely adopted in HR analytics.

And once you know who is likely to leave, you actually have levers to pull: mentorship programs, targeted retention bonuses, workload redistribution, or career development opportunities. Importantly, these are one-time trigger interventions—not endless nudges that require a continuous AI babysitter. That makes them far easier to implement and measure.

The lesson: Don’t just bolt AI onto a problem because it sounds futuristic. Ask yourself:

👉 What’s the question AI can answer that directly drives ROI?

That’s where the value lives.

What business question have you helped reframe in your work? Drop me a line and let me know!

Greg

P.S. For further reading, also check out point 4: Make sure your AI model is answering the right question here: https://www.uxforai.com/p/stop-f-cking-up-ai-projects-avoid-these-5-pitfalls

P.P.S. What did you think of this #AISnowflake? Want more of those on your nose?